No changes on our side before tonight. But we've set the minimum GPU ram to 2048mb, and I see your GTX reports `NVIDIA GeForce GTX 750 Ti (1994MB) driver: 455.38 OpenCL: 1.2`. 1994 < 2048, so you're probably being filtered out because of that.

I just reset the plan class to a minimum of 1900mb. We'll likely still be OK, but I suspect it's really pushing it.

Thank you. We will find out soon enough if I am biting more than I can chew...a task is currently downloading.

The task finished without a problem. :-)

AMD Rocm 4.0 published

Yay.. noted. I hadn't seen that yet.

I'll recompile/update/retest/re-release the ROCM client in the next week or so.

Spent some time trying to make this work yesterday. I looks like there's still some issues with pytorch and ROCm 4.0 at the moment. It's a little frustrating, because pytorch itself seems to compile fine, but when you use the pytorch interface to query GPU information it can't find header files it had no problem finding in rocm 3.9. I'll continue to poke at it, but it might not be ready as quickly as I'd hoped.

ROCm-3.10 might be a better option, and that should work with POLARIS as well as VEGA at least.

Hello,
due to the second GPU (old) I have for display but not for crunching, I have to only get the maximum driver version for that old card.
390.143 is still supported by NVIDIA, and I am using it with my GTX 750 Ti, would it be possible to allow my GPU to help this project by allowing this version of NVIDIA driver?

That's an interesting question! The short answer is probably not, since we're using pytorch and that requires cuda 10, which requires a minimum driver version in of 440(?) or higher. But I'll look into it.

Thanks :)
Indeed, I am using CUDA 9.1. Maybe the solution is to change hardware one day!